Zoe Hu 1 Tamas Ungi 2 Jay Engel 1 Gabor Fichtinger 2 Doris Jabs 1 1 School of Medicine Queens University 2 School of Computing Queens University No conflicts of interest to disclose ID: 1041107
Download Presentation The PPT/PDF document "Deep Learning-Based Automatic Tumour Seg..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
1. Deep Learning-Based Automatic Tumour Segmentation in Breast-Conserving Surgery Navigation SystemsZoe Hu1, Tamas Ungi2, Jay Engel1, Gabor Fichtinger2, Doris Jabs11School of Medicine, Queen’s University2School of Computing, Queen’s University
2. No conflicts of interest to discloseStudy approved by the Queen’s University health Sciences and Affiliated Teaching Hospitals Research Ethics Board- 2 -
3. Breast-Conserving Surgery- 3 -30% of patients undergoing BCS will have positive margins on post-op pathology analysis
4. NaviKnife
5. ObjectiveDevelop intraoperative automatic segmentation of the breast tumour on 3D ultrasound imaging to replace manual contouring by a radiologist.- 5 -
6. U-Net
7. U-Net Training- 7 -
8. Implementation StrategiesHyperparameter OptimizationRandom search Baseline Trial and error Model parameters - 8 -
9. Implementation StrategiesWeighted categorical loss functionHealthy adipose tissue >> Tumour tissuePredict only healthy tissue Specificity 80%Ideal ratio = 85:15- 9 -80%20%
10. Implementation StrategiesData augmentationTranslation, zoom, shift, rotation, flip- 10 -
11. ModelModel ParameterValueLayers7Kernels3x3 downsampling4x4 upsamplingLearning rate1e-4 Loss functionWeighted categorical loss fxnActivation functionSoftmaxBatch size32Epochs200- 11 -
12. Method: Cross-Validation80% for training, 20% for testing Prevents Overfitting- 12 -TrainTrainTrainTrainTestTrainTrainTrainTrainTestTrainTrainTrainTrainTestTrainTrainTrainTrainTestTrainTrainTrainTrainTest1.2.3.4.5.Model “blinded” to test set
13. Method: Evaluation- 13 -
14. ResultsAccuracy MetricValueArea under the ROC curve (AUC)0.94Dice similarity coefficient (DSC)0.70Sensitivity92%Specificity65%- 14 -Harmonic mean of sensitivity and specificityCapability of the model in distinguishing the classes
15. Existing WorkComparisons to similar studies:- 15 -ModelAUCZhuang et al.0.92Byra et al.0.95Almajalid et al.0.82Wang et al.0.92Our model0.94
16. Example 1
17. Example 2
18. Clinical RelevanceSurvey results:100% of responses rated tumour contour quality in 2D and 3D above 70%78% of responses rated tumour contour quality in 2D and 3D above 80%56% of responders stated that they would be comfortable using the automatic tumour contours in breast conserving surgery- 18 -
19. ConclusionHigh sensitivity and AUC valuesGood visual representation and robust 3D reconstruction pipelineSurvey results positive for contour quality- 19 -
20. Next StepsnnU-Net
21. Thank YouQueen’s Perk Lab: Dr. Tamas Ungi Dr. Gabor FichtingerKingston Health Sciences Center: Dr. Jay Engel Dr. Doris Jabs- 21 -
22. ReferencesCanada, P. H. A. of. (2019, December 9). Breast Cancer [Education and awareness]. Aem. https://www.canada.ca/en/public-health/services/chronic-diseases/cancer/breast-cancer.htmlCao, Z., Duan, L., Yang, G., Yue, T., Chen, Q., Fu, H., & Xu, Y. (2017). Breast Tumor Detection in Ultrasound Images Using Deep Learning. In G. Wu, B. C. Munsell, Y. Zhan, W. Bai, G. Sanroma, & P. Coupé (Eds.), Patch-Based Techniques in Medical Imaging (pp. 121–128). Springer International Publishing. https://doi.org/10.1007/978-3-319-67434-6_14Chen, K., Li, S., Li, Q., Zhu, L., Liu, Y., Song, E., & Su, F. (2016). Breast-conserving Surgery Rates in Breast Cancer Patients With Different Molecular Subtypes. Medicine, 95(8). https://doi.org/10.1097/MD.0000000000002593Deep Learning in Medical Image Analysis. (n.d.). Retrieved March 6, 2020, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5479722/Dua, S. M., Gray, R. J., & Keshtgar, M. (2011). Strategies for localisation of impalpable breast lesions. Breast (Edinburgh, Scotland), 20(3), 246–253. https://doi.org/10.1016/j.breast.2011.01.007Fajdic, J., Djurovic, D., Gotovac, N., & Hrgovic, Z. (2013). Criteria and Procedures for Breast Conserving Surgery. Acta Informatica Medica, 21(1), 16–19. https://doi.org/10.5455/AIM.2013.21.16-19Gauvin, G., Yeo, C. T., Ungi, T., Merchant, S., Lasso, A., Jabs, D., Vaughan, T., Rudan, J. F., Walker, R., Fichtinger, G., & Engel, C. J. (2019). Real-time electromagnetic navigation for breast-conserving surgery using NaviKnife technology: A matched case-control study. The Breast Journal. https://doi.org/10.1111/tbj.13480Hargreaves, A. C., Mohamed, M., & Audisio, R. A. (2014). Intra-operative guidance: Methods for achieving negative margins in breast conserving surgery. Journal of Surgical Oncology, 110(1), 21–25. https://doi.org/10.1002/jso.23645Klarenbach, S., Sims-Jones, N., Lewin, G., Singh, H., Thériault, G., Tonelli, M., Doull, M., Courage, S., Garcia, A. J., Thombs, B. D., & Canadian Task Force on Preventive Health Care. (2018). Recommendations on screening for breast cancer in women aged 40-74 years who are not at increased risk for breast cancer. CMAJ: Canadian Medical Association Journal = Journal de l’Association Medicale Canadienne, 190(49), E1441–E1451. https://doi.org/10.1503/cmaj.180463Pan, H., Wu, N., Ding, H., Ding, Q., Dai, J., Ling, L., Chen, L., Zha, X., Liu, X., Zhou, W., & Wang, S. (2013). Intraoperative ultrasound guidance is associated with clear lumpectomy margins for breast cancer: A systematic review and meta-analysis. PloS One, 8(9), e74028. https://doi.org/10.1371/journal.pone.0074028Ronneberger, O., Fischer, P., & Brox, T. (2015). U-Net: Convolutional Networks for Biomedical Image Segmentation. In N. Navab, J. Hornegger, W. M. Wells, & A. F. Frangi (Eds.), Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015 (pp. 234–241). Springer International Publishing. https://doi.org/10.1007/978-3-319-24574-4_28Wood, W. C. (2013). Close/positive margins after breast-conserving therapy: Additional resection or no resection? Breast (Edinburgh, Scotland), 22 Suppl 2, S115-117. https://doi.org/10.1016/j.breast.2013.07.022Zeimarani, B., Costa, M. G. F., Nurani, N. Z., & Costa Filho, C. F. F. (2019). A Novel Breast Tumor Classification in Ultrasound Images, Using Deep Convolutional Neural Network. In R. Costa-Felix, J. C. Machado, & A. V. Alvarenga (Eds.), XXVI Brazilian Congress on Biomedical Engineering (pp. 89–94). Springer. https://doi.org/10.1007/978-981-13-2517-5_14- 22 -